This work is focused on designing and implementing a real-time video-based face identification system with low memory and computational requirements and high recognition rates. Since pro le features are stronger and, therefore, better when characterising faces than frontal faces, the system will detect and identify not only pure frontal but also pro le faces. This property of pro le faces will help to improve face recognition rates depending on the strategy for fusion of results used. Also, dimensionality reduction techniques will be studied and tested in order to nd the fastest and most efective one. Modi cation in k Nearest Neighbor classi er will be carried out to add a penalisation factor in function of the distance, increasing classi cation results and strictness. In order to nd which are the best options for reducing computational requirements in a face identi cation system several simulations will be performed. Among many others, simulations will look for optimal values of the k parameter in k Nearest Neighbor, the number of transformed coe cients kept in a feature vector or the minimum size of face images and will test dimensionality reduction in images, variation of the number of models or fusion of results. Finally, this work will show how a real-time system can be implemented in an ordinary computer obtaining successful results whether it be in real-time, adverse or controlled conditions environments.